Millions of Hack Attempts Hit PHP Bugs Within Hours—and AI Is Why

Millions of exploitation attempts slammed into newly disclosed PHP vulnerabilities within hours of their publication, according to an analysis from Cloudflare. That is the hardest number in this story, and it frames everything that followed: criminal hackers moved on the flaws almost instantly, and the security world spent the rest of the week asking whether AI just changed the economics of exploitation.

So far, the worst-case outcome hasn't materialized. The PHP Foundation says it hasn't seen any successful exploitation of the bugs, which researchers collectively dubbed PHP2Shell. Many of the potential exploits circulating publicly appear to be ineffective. Web admins and security teams scrambled to patch, and the patches largely held.

But the deeper story isn't about the attack attempts. It's about who found the bugs, how, and why now. As Lily Hay Newman and Matt Burgess report in this week's edition of Kernel Panic, their weekly newsletter on privacy and digital security from WIRED, advanced AI models helped researchers uncover practical exploitation paths for vulnerabilities that had been known, at least in part, for years. AI helped turn a historic vulnerability into a current crisis.

What exactly happened?

The timeline matters here. Researchers at the security company Calif were the first to identify the PHP2Shell vulnerabilities and build working exploits. They did it in early August. Calif's work partially used frontier AI tools as part of OpenAI's Patch the Planet program. But the researchers also relied on expert human knowledge of PHP, its dynamics, and its specific weaknesses. The machines didn't do this alone.

Some of the underlying issues had been identified in 2024. At the time, they were deemed too complex to exploit, and so they sat. That decision was rational by the standards of the day. A bug you can't practically weaponize ranks below the ones you can.

After Calif's discovery, researchers at OpenAI and at Wiz, the security firm owned by Google, independently discovered PHP2Shell flaws using AI tools. The parallel findings underscored the point: the same class of tooling, pointed at the same codebase, reached the same conclusion.

Wiz security researcher Sagi Tzadik put the dynamic plainly: "The vulnerability had been known to the PHP team for years, but because it could not be leveraged by an attacker—at least, not a human one—it was not addressed."

That quote is the hinge of the whole story. An entire category of software flaws existed in a kind of suspended animation—real, documented, but practically unreachable. AI tools reached them.

Does this mean a wave of decade-old bugs is coming?

Maybe, but probably not an endless one. No software is perfect, and even the most critical, foundational digital tools carry inevitable flaws. The expanding capabilities of AI models have made it faster to analyze vulnerabilities and, potentially, to exploit them. That speed lands hardest on under-resourced IT teams, developers, and admins who now have to keep up with disclosure cycles that compress from months to days.

The broader environment is already strained. Microsoft's security updates this month alone included a record 1,453 security vulnerabilities. Bank of America, meanwhile, recently hired 1,000 consultants to fix flaws discovered by Anthropic's Mythos, according to the newsletter. Those two data points together sketch the new baseline: more findings, more fixes, more pressure.

Security researchers are quick to add a caveat, though. AI is not doing anything fundamentally new. It magnifies existing dynamics of vulnerability discovery and coordinated disclosure—faster analysis, faster exploit development, faster patching, faster attacks.

Jake Williams, vice president of research and development at Hunter Strategy and a former NSA hacker, sees a limited supply of targets like PHP. "We'll definitely see more of these long-life bugs that are in every version of the software for the last decade or whatever," he said. "But I think there's only so much of this low-hanging fruit. There isn't much else that has the specific reach of PHP. It's a pretty wild attack surface—most other tools aren't exposed to the internet by design."

Williams's point about reach deserves emphasis. PHP powers an enormous slice of the public web, and it is exposed to the internet by design. Most other foundational tools simply aren't sitting on the network edge in the same way. A bug in PHP is a bug in the front door of millions of sites.

Where do humans still fit?

Everywhere, according to the people who did the work. Researchers note that even with frontier models, most examples of sophisticated AI-aided hacking also involve substantial human expertise. The tools accelerate experts. They don't replace them.

Ryan Austin, the Calif researcher who first discovered the PHP flaws, wrote this week: "No model is simply capable of unearthing 'all bugs' with simple prompts. At least for now." The qualifier at the end—"at least for now"—carries more weight than the rest of the sentence.

That three-word hedge is where the honest uncertainty lives. Today's models need skilled teams to produce results like PHP2Shell. Nobody in this story claims that will remain true indefinitely, and nobody claims to know when it changes.

Security engineer and researcher Niels Provos is more sanguine than most. "To be honest, I am not really worried," he said, about the prospect of AI finding an overwhelming number of historic bugs that are newly exploitable, or of new vulnerabilities that couldn't have been practically exploited without AI. "People have now been looking with AI for over a year, and it required a highly skilled team to find this one. I guess we will see whether I am completely wrong, but I doubt it."

His evidence is empirical rather than theoretical. The AI-assisted bug-hunting effort has been running for more than a year. In that time, the headline result is one major vulnerability cluster, found by experts, patched before mass compromise.

Why does this story matter beyond PHP?

Because it establishes a pattern. A known flaw, judged unexploitable by human standards, became exploitable when AI tools changed the cost calculation. Attackers then generated millions of exploitation attempts within hours, per Cloudflare. Defenders held the line this time, but the structural shift is real: the shelf life of "too hard to exploit" just got shorter.

The stakes run in both directions. The same AI capabilities that helped Calif, OpenAI, and Wiz find PHP2Shell are available to attackers, and the Cloudflare data shows they're using them fast. Coordinated disclosure—getting patches out before exploits spread—now has to outrun automated exploit development.

For the organizations running PHP at scale, the immediate task is simple: confirm the patches are applied, because the PHP Foundation's clean record holds only as long as systems stay current. For everyone else, the PHP2Shell episode is the first clear demonstration of a dynamic security teams will now budget for—old bugs, new math, and a shrinking window between disclosure and attack.